Artificial neural circuit using pulse coding

Abstract

Disclosed is an artificial neural circuit using a pulse coding such as a stochastic pulse coding or a noise feedback pulse coding, the circuit comprising a synapse circuit section for producing and absorbing a current signal proportional to a weight voltage signal upon an externally applied current signal being supplied; a neuron body circuit section for spacio-temporally integrating output signal of the synapse circuit section to produce an analog voltage signal; and an axon hillock circuit for converting the analog voltage signal into a pulse train using a predetermined reference signal. The synapse circuit section including a first input terminal for receiving a first reference voltage; a second input terminal for receiving a second reference voltage; a third input terminal for receiving a third reference voltage; a fourth terminal for receiving the weight voltage signal; a first transistor having drain and source connected to the first reference voltage and a junction point, respectively, and gate connected to the fourth input terminal; a second transistor having drain and source connected to the junction point and the second input terminal, respectively, and gate connected to the third input terminal; and a third transistor having drain and source connected respectively to the junction point and an output terminal of the synapse circuit section, and gate connected to an input terminal for receiving the externally applied current signal.

What is claimed is: 
     
       1. An artificial neural circuit in which an operating principal of a biological neuron is embodied, said circuit comprising: a synapse circuit section for producing and absorbing a current signal proportional to a weight voltage signal upon an externally applied current signal being supplied;   said synapse circuit section including a first input terminal for receiving a first reference voltage; a second input terminal for receiving a second reference voltage; a third input terminal for receiving a third reference voltage; a fourth input terminal for receiving the weight voltage signal; a first transistor having drain and source connected to the first reference voltage and a junction point, respectively, and gate connected to the fourth input terminal; a second transistor having drain and source connected to the junction point and the second input terminal, respectively, and gate connected to the third input terminal; and a third transistor having drain and source connected respectively to the junction point and an output terminal of the synapse circuit section, and gate connected to an input terminal for receiving the externally applied current signal;   a neuron body circuit section for spacio-temporally integrating output signal of the synapse circuit section to produce an analog voltage signal; and   an axon hillock circuit for converting the analog voltage signal into a pulse train using a predetermined reference signal.   
     
     
       2. The artificial neural circuit according to claim 1, wherein said axon hillock circuit comprises a random wave generator for generating the predetermined reference signal, a first comparator for comparing the predetermined reference signal with an output signal of the neuron body circuit section, and an AND gate for receiving an output signal of the first comparator and a predetermined frequency signal, so as to generate the pulse train having one half of duty ratio of the output signal of the first comparator. 
     
     
       3. The artificial neural circuit according to claim 1, wherein said axon hillock circuit section comprises integrating means for integrating the output signal of the neuron body circuit section to produce an integrated signal, a second comparator for comparing the integrated signal with a threshold signal to produce a resulting signal, duty converting means for generating an output signal having one half of duty ratio of the resulting signal of the second comparator, and feedback means in response to the output signal of the duty converting means to supply a reference signal for the integrating means. 
     
     
       4. The artificial neural circuit according to claim 1, wherein said neuron body circuit section comprises an operational amplifier having an inverting terminal connected with the source of the second transistor and a non-inverting terminal connected with a ground, a capacitor connected between the inverting terminal and an output terminal of the operational amplifier, and a resistor connected in parallel to the capacitor, so that integration time of the neuron body circuit section is controlled dependent upon time constant of the capacitor and resistor.